摘要
行人再识别是当前计算机视觉的研究难点和热点问题。根据行人再识别技术最新研究动态,对其研究现状进行了系统梳理,重点分析了特征学习和分类器算法两个最为核心的问题。首先将行人再识别的研究方法进行分类,其次整理了相关的研究资源,然后将特征学习分类为局部特征、块特征和全局特征,分类器算法分类为基于图像的建模方法、基于视频的建模方法、End-to-End的研究方法和深度学习模型,最后总结了行人再识别技术存在的主要问题,并对行人再识别检测技术的研究趋势进行展望。
Person Re-Identification is one of the hot and knotty issues in computer vision.This study conducts a detailed survey on state-of-the-art Person Re-Identification methods according to the latest development trend,focusing on the two most important problems,they are feature learning and classifier algorithms.First,we divide these methods into different categories,reorganizes relevant research materials,feature learning are divided into three subcategories,they are local features,patch features and global features,classifier algorithms are divided into four subcategories,they are image-based learning methods,video-based learning methods,End-to-End research methods and deep learning.Second,the key issues existed in Person Re-Identification are summarized.Finally,some future research trends are proposed.
作者
赵翎妗
郭方
夏道勋
ZHAO Lingjin;GUO Fang;XIA Daoxun(School of Big Data and Computer Science,Guizhou Normal University,Guiyang,Guizhou 550025,China;The EngineeringLaboratory for Applied Technology of Big Data in Education in Guizhou Province Guizhou Normal University,Guiyang,Guizhou 550025,China)
出处
《贵州师范大学学报（自然科学版）》
CAS
2019年第6期114-122,共9页
Journal of Guizhou Normal University：Natural Sciences
基金
国家自然科学基金项目(61762023)
贵州省自然科学基金项目(黔科合LH字[2015]7784号)
贵州师范大学2017年博士科研启动项目
贵州省科技厅新苗培育专项(黔科合平台人才[2017]5726)
关键词
行人再识别
目标检测
度量学习
深度学习
综述
person re-identification
object detection
metric learning
deep learning
survey
